[ { "path": ".gitignore", "content": "# Add by user\n.vscode/\nresult/\ndata/\nsave_model/\n\n# Byte-compiled / optimized / DLL files\n__pycache__/\n*.py[cod]\n*$py.class\n\n# C extensions\n*.so\n\n# Distribution / packaging\n.Python\nbuild/\ndevelop-eggs/\ndist/\ndownloads/\neggs/\n.eggs/\nlib/\nlib64/\nparts/\nsdist/\nvar/\nwheels/\n*.egg-info/\n.installed.cfg\n*.egg\nMANIFEST\n\n# PyInstaller\n# Usually these files are written by a python script from a template\n# before PyInstaller builds the exe, so as to inject date/other infos into it.\n*.manifest\n*.spec\n\n# Installer logs\npip-log.txt\npip-delete-this-directory.txt\n\n# Unit test / coverage reports\nhtmlcov/\n.tox/\n.nox/\n.coverage\n.coverage.*\n.cache\nnosetests.xml\ncoverage.xml\n*.cover\n.hypothesis/\n.pytest_cache/\n\n# Translations\n*.mo\n*.pot\n\n# Django stuff:\n*.log\nlocal_settings.py\ndb.sqlite3\n\n# Flask stuff:\ninstance/\n.webassets-cache\n\n# Scrapy stuff:\n.scrapy\n\n# Sphinx documentation\ndocs/_build/\n\n# PyBuilder\ntarget/\n\n# Jupyter Notebook\n.ipynb_checkpoints\n\n# IPython\nprofile_default/\nipython_config.py\n\n# pyenv\n.python-version\n\n# celery beat schedule file\ncelerybeat-schedule\n\n# SageMath parsed files\n*.sage.py\n\n# Environments\n.env\n.venv\nenv/\nvenv/\nENV/\nenv.bak/\nvenv.bak/\n\n# Spyder project settings\n.spyderproject\n.spyproject\n\n# Rope project settings\n.ropeproject\n\n# mkdocs documentation\n/site\n\n# mypy\n.mypy_cache/\n.dmypy.json\ndmypy.json" }, { "path": "LICENSE", "content": "MIT License\n\nCopyright (c) 2018 IDKiro\n\nPermission is hereby granted, free of charge, to any person obtaining a copy\nof this software and associated documentation files (the \"Software\"), to deal\nin the Software without restriction, including without limitation the rights\nto use, copy, modify, merge, publish, distribute, sublicense, and/or sell\ncopies of the Software, and to permit persons to whom the Software is\nfurnished to do so, subject to the following conditions:\n\nThe above copyright notice and this permission notice shall be included in all\ncopies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\nIMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\nFITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\nAUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\nLIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\nOUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE\nSOFTWARE.\n" }, { "path": "README.md", "content": "# CBDNet-pytorch\n\nIt's an unofficial PyTorch implementation of CBDNet.\n\nWe used higher quality real and synthetic datasets for training and achieved better performance on DND.\n\n[CBDNet in MATLAB](https://github.com/GuoShi28/CBDNet)\n\n[CBDNet in Tensorflow](https://github.com/IDKiro/CBDNet-tensorflow)\n\n## Quick Start\n\nDownload the dataset and pretrained model from [GoogleDrive](https://drive.google.com/drive/folders/1-e2nPCr_eP1cTDhFFes27Rjj-QXzMk5u?usp=sharing).\n\nExtract the files to `data` folder and `save_model` folder as follow:\n\n```\n~/\n data/\n SIDD_train/\n ... (scene id)\n Syn_train/\n ... (id)\n DND/\n images_srgb/\n ... (mat files)\n ... (mat files)\n save_model/\n checkpoint.pth.tar\n```\n\nTrain the model:\n\n```\npython train.py\n```\n\nPredict using the trained model:\n\n```\npython predict.py input_filename output_filename\n```\n\n## Network Structure\n\n![Image of Network](imgs/CBDNet_v13.png)\n\n## Realistic Noise Model\nGiven a clean image `x`, the realistic noise model can be represented as:\n\n![](http://latex.codecogs.com/gif.latex?\\\\textbf{y}=f(\\\\textbf{DM}(\\\\textbf{L}+n(\\\\textbf{L}))))\n\n![](http://latex.codecogs.com/gif.latex?n(\\\\textbf{L})=n_s(\\\\textbf{L})+n_c)\n\nWhere `y` is the noisy image, `f(.)` is the CRF function and the irradiance ![](http://latex.codecogs.com/gif.latex?\\\\textbf{L}=\\\\textbf{M}f^{-1}(\\\\textbf{x})) , `M(.)` represents the function that convert sRGB image to Bayer image and `DM(.)` represents the demosaicing function.\n\nIf considering denosing on compressed images, \n\n![](http://latex.codecogs.com/gif.latex?\\\\textbf{y}=JPEG(f(\\\\textbf{DM}(\\\\textbf{L}+n(\\\\textbf{L})))))\n\n## Result\n\n![](imgs/results.png)\n" }, { "path": "dataset/__init__.py", "content": "" }, { "path": "dataset/loader.py", "content": "import os\nimport random\nimport torch\nimport numpy as np\nimport glob\nfrom torch.utils.data import Dataset\n\nfrom utils import read_img, hwc_to_chw\n\n\ndef get_patch(imgs, patch_size):\n\tH = imgs[0].shape[0]\n\tW = imgs[0].shape[1]\n\n\tps_temp = min(H, W, patch_size)\n\n\txx = np.random.randint(0, W-ps_temp) if W > ps_temp else 0\n\tyy = np.random.randint(0, H-ps_temp) if H > ps_temp else 0\n\n\tfor i in range(len(imgs)):\n\t\timgs[i] = imgs[i][yy:yy+ps_temp, xx:xx+ps_temp, :]\n\n\tif np.random.randint(2, size=1)[0] == 1:\n\t\tfor i in range(len(imgs)):\n\t\t\timgs[i] = np.flip(imgs[i], axis=1)\n\tif np.random.randint(2, size=1)[0] == 1: \n\t\tfor i in range(len(imgs)):\n\t\t\timgs[i] = np.flip(imgs[i], axis=0)\n\tif np.random.randint(2, size=1)[0] == 1:\n\t\tfor i in range(len(imgs)):\n\t\t\timgs[i] = np.transpose(imgs[i], (1, 0, 2))\n\n\treturn imgs\n\n\nclass Real(Dataset):\n\tdef __init__(self, root_dir, sample_num, patch_size=128):\n\t\tself.patch_size = patch_size\n\n\t\tfolders = glob.glob(root_dir + '/*')\n\t\tfolders.sort()\n\n\t\tself.clean_fns = [None] * sample_num\n\t\tfor i in range(sample_num):\n\t\t\tself.clean_fns[i] = []\n\n\t\tfor ind, folder in enumerate(folders):\n\t\t\tclean_imgs = glob.glob(folder + '/*GT_SRGB*')\n\t\t\tclean_imgs.sort()\n\n\t\t\tfor clean_img in clean_imgs:\n\t\t\t\tself.clean_fns[ind % sample_num].append(clean_img)\n\n\tdef __len__(self):\n\t\tl = len(self.clean_fns)\n\t\treturn l\n\n\tdef __getitem__(self, idx):\n\t\tclean_fn = random.choice(self.clean_fns[idx])\n\n\t\tclean_img = read_img(clean_fn)\n\t\tnoise_img = read_img(clean_fn.replace('GT_SRGB', 'NOISY_SRGB'))\n\n\t\tif self.patch_size > 0:\n\t\t\t[clean_img, noise_img] = get_patch([clean_img, noise_img], self.patch_size)\n\n\t\treturn hwc_to_chw(noise_img), hwc_to_chw(clean_img), np.zeros((3, self.patch_size, self.patch_size)), np.zeros((3, self.patch_size, self.patch_size))\n\n\nclass Syn(Dataset):\n\tdef __init__(self, root_dir, sample_num, patch_size=128):\n\t\tself.patch_size = patch_size\n\n\t\tfolders = glob.glob(root_dir + '/*')\n\t\tfolders.sort()\n\n\t\tself.clean_fns = [None] * sample_num\n\t\tfor i in range(sample_num):\n\t\t\tself.clean_fns[i] = []\n\n\t\tfor ind, folder in enumerate(folders):\n\t\t\tclean_imgs = glob.glob(folder + '/*GT_SRGB*')\n\t\t\tclean_imgs.sort()\n\n\t\t\tfor clean_img in clean_imgs:\n\t\t\t\tself.clean_fns[ind % sample_num].append(clean_img)\n\n\tdef __len__(self):\n\t\tl = len(self.clean_fns)\n\t\treturn l\n\n\tdef __getitem__(self, idx):\n\t\tclean_fn = random.choice(self.clean_fns[idx])\n\n\t\tclean_img = read_img(clean_fn)\n\t\tnoise_img = read_img(clean_fn.replace('GT_SRGB', 'NOISY_SRGB'))\n\t\tsigma_img = read_img(clean_fn.replace('GT_SRGB', 'SIGMA_SRGB')) / 15.\t# inverse scaling\n\n\t\tif self.patch_size > 0:\n\t\t\t[clean_img, noise_img, sigma_img] = get_patch([clean_img, noise_img, sigma_img], self.patch_size)\n\n\t\treturn hwc_to_chw(noise_img), hwc_to_chw(clean_img), hwc_to_chw(sigma_img), np.ones((3, self.patch_size, self.patch_size))" }, { "path": "model/__init__.py", "content": "" }, { "path": "model/cbdnet.py", "content": "\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\n\n\nclass single_conv(nn.Module):\n def __init__(self, in_ch, out_ch):\n super(single_conv, self).__init__()\n self.conv = nn.Sequential(\n nn.Conv2d(in_ch, out_ch, 3, padding=1),\n nn.ReLU(inplace=True)\n )\n\n def forward(self, x):\n return self.conv(x)\n\n\nclass up(nn.Module):\n def __init__(self, in_ch):\n super(up, self).__init__()\n self.up = nn.ConvTranspose2d(in_ch, in_ch//2, 2, stride=2)\n\n def forward(self, x1, x2):\n x1 = self.up(x1)\n \n # input is CHW\n diffY = x2.size()[2] - x1.size()[2]\n diffX = x2.size()[3] - x1.size()[3]\n\n x1 = F.pad(x1, (diffX // 2, diffX - diffX//2,\n diffY // 2, diffY - diffY//2))\n\n x = x2 + x1\n return x\n\n\nclass outconv(nn.Module):\n def __init__(self, in_ch, out_ch):\n super(outconv, self).__init__()\n self.conv = nn.Conv2d(in_ch, out_ch, 1)\n\n def forward(self, x):\n x = self.conv(x)\n return x\n\n\nclass FCN(nn.Module):\n def __init__(self):\n super(FCN, self).__init__()\n self.fcn = nn.Sequential(\n nn.Conv2d(3, 32, 3, padding=1),\n nn.ReLU(inplace=True),\n nn.Conv2d(32, 32, 3, padding=1),\n nn.ReLU(inplace=True),\n nn.Conv2d(32, 32, 3, padding=1),\n nn.ReLU(inplace=True),\n nn.Conv2d(32, 32, 3, padding=1),\n nn.ReLU(inplace=True),\n nn.Conv2d(32, 3, 3, padding=1),\n nn.ReLU(inplace=True)\n )\n \n def forward(self, x):\n return self.fcn(x)\n\n\nclass UNet(nn.Module):\n def __init__(self):\n super(UNet, self).__init__()\n \n self.inc = nn.Sequential(\n single_conv(6, 64),\n single_conv(64, 64)\n )\n\n self.down1 = nn.AvgPool2d(2)\n self.conv1 = nn.Sequential(\n single_conv(64, 128),\n single_conv(128, 128),\n single_conv(128, 128)\n )\n\n self.down2 = nn.AvgPool2d(2)\n self.conv2 = nn.Sequential(\n single_conv(128, 256),\n single_conv(256, 256),\n single_conv(256, 256),\n single_conv(256, 256),\n single_conv(256, 256),\n single_conv(256, 256)\n )\n\n self.up1 = up(256)\n self.conv3 = nn.Sequential(\n single_conv(128, 128),\n single_conv(128, 128),\n single_conv(128, 128)\n )\n\n self.up2 = up(128)\n self.conv4 = nn.Sequential(\n single_conv(64, 64),\n single_conv(64, 64)\n )\n\n self.outc = outconv(64, 3)\n\n def forward(self, x):\n inx = self.inc(x)\n\n down1 = self.down1(inx)\n conv1 = self.conv1(down1)\n\n down2 = self.down2(conv1)\n conv2 = self.conv2(down2)\n\n up1 = self.up1(conv2, conv1)\n conv3 = self.conv3(up1)\n\n up2 = self.up2(conv3, inx)\n conv4 = self.conv4(up2)\n\n out = self.outc(conv4)\n return out\n\n\nclass Network(nn.Module):\n def __init__(self):\n super(Network, self).__init__()\n self.fcn = FCN()\n self.unet = UNet()\n \n def forward(self, x):\n noise_level = self.fcn(x)\n concat_img = torch.cat([x, noise_level], dim=1)\n out = self.unet(concat_img) + x\n return noise_level, out\n\n\nclass fixed_loss(nn.Module):\n def __init__(self):\n super().__init__()\n \n def forward(self, out_image, gt_image, est_noise, gt_noise, if_asym):\n l2_loss = F.mse_loss(out_image, gt_image)\n\n asym_loss = torch.mean(if_asym * torch.abs(0.3 - torch.lt(gt_noise, est_noise).float()) * torch.pow(est_noise - gt_noise, 2))\n\n h_x = est_noise.size()[2]\n w_x = est_noise.size()[3]\n count_h = self._tensor_size(est_noise[:, :, 1:, :])\n count_w = self._tensor_size(est_noise[:, :, : ,1:])\n h_tv = torch.pow((est_noise[:, :, 1:, :] - est_noise[:, :, :h_x-1, :]), 2).sum()\n w_tv = torch.pow((est_noise[:, :, :, 1:] - est_noise[:, :, :, :w_x-1]), 2).sum()\n tvloss = h_tv / count_h + w_tv / count_w\n\n loss = l2_loss + 0.5 * asym_loss + 0.05 * tvloss\n\n return loss\n\n def _tensor_size(self, t):\n return t.size()[1]*t.size()[2]*t.size()[3]" }, { "path": "predict.py", "content": "import os, time, scipy.io, shutil\nimport numpy as np\nimport torch\nimport torch.nn as nn\nimport argparse\nimport cv2\n\nfrom model.cbdnet import Network\nfrom utils import read_img, chw_to_hwc, hwc_to_chw\n\nparser = argparse.ArgumentParser(description = 'Test')\nparser.add_argument('input_filename', type=str)\nparser.add_argument('output_filename', type=str)\nargs = parser.parse_args()\n\nsave_dir = './save_model/'\n\nmodel = Network()\nmodel.cuda()\nmodel = nn.DataParallel(model)\n\nmodel.eval()\n\nif os.path.exists(os.path.join(save_dir, 'checkpoint.pth.tar')):\n # load existing model\n model_info = torch.load(os.path.join(save_dir, 'checkpoint.pth.tar'))\n model.load_state_dict(model_info['state_dict'])\nelse:\n print('Error: no trained model detected!')\n exit(1)\n\ninput_image = read_img(args.input_filename)\ninput_var = torch.from_numpy(hwc_to_chw(input_image)).unsqueeze(0).cuda()\n\nwith torch.no_grad():\n _, output = model(input_var)\n\noutput_image = chw_to_hwc(output[0,...].cpu().numpy())\noutput_image = np.uint8(np.round(np.clip(output_image, 0, 1) * 255.))[: ,: ,::-1]\n\ncv2.imwrite(args.output_filename, output_image)" }, { "path": "train.py", "content": "import os, time, shutil\nimport argparse\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\n\nfrom utils import AverageMeter\nfrom dataset.loader import Real, Syn\nfrom model.cbdnet import Network, fixed_loss\n\n\nparser = argparse.ArgumentParser(description = 'Train')\nparser.add_argument('--bs', default=32, type=int, help='batch size')\nparser.add_argument('--ps', default=128, type=int, help='patch size')\nparser.add_argument('--lr', default=2e-4, type=float, help='learning rate')\nparser.add_argument('--epochs', default=5000, type=int, help='sum of epochs')\nargs = parser.parse_args()\n\n\ndef train(train_loader, model, criterion, optimizer):\n\tlosses = AverageMeter()\n\tmodel.train()\n\n\tfor (noise_img, clean_img, sigma_img, flag) in train_loader:\n\t\tinput_var = noise_img.cuda()\n\t\ttarget_var = clean_img.cuda()\n\t\tsigma_var = sigma_img.cuda()\n\t\tflag_var = flag.cuda()\n\n\t\tnoise_level_est, output = model(input_var)\n\n\t\tloss = criterion(output, target_var, noise_level_est, sigma_var, flag_var)\n\t\tlosses.update(loss.item())\n\n\t\toptimizer.zero_grad()\n\t\tloss.backward()\n\t\toptimizer.step()\n\t\n\treturn losses.avg\n\n\nif __name__ == '__main__':\n\tsave_dir = './save_model/'\n\n\tmodel = Network()\n\tmodel.cuda()\n\tmodel = nn.DataParallel(model)\n\n\tif os.path.exists(os.path.join(save_dir, 'checkpoint.pth.tar')):\n\t\t# load existing model\n\t\tmodel_info = torch.load(os.path.join(save_dir, 'checkpoint.pth.tar'))\n\t\tprint('==> loading existing model:', os.path.join(save_dir, 'checkpoint.pth.tar'))\n\t\tmodel.load_state_dict(model_info['state_dict'])\n\t\toptimizer = torch.optim.Adam(model.parameters())\n\t\toptimizer.load_state_dict(model_info['optimizer'])\n\t\tscheduler = torch.optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=args.epochs)\n\t\tscheduler.load_state_dict(model_info['scheduler'])\n\t\tcur_epoch = model_info['epoch']\n\telse:\n\t\tif not os.path.isdir(save_dir):\n\t\t\tos.makedirs(save_dir)\n\t\t# create model\n\t\toptimizer = torch.optim.Adam(model.parameters(), lr=args.lr)\n\t\tscheduler = torch.optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=args.epochs)\n\t\tcur_epoch = 0\n\t\t\n\tcriterion = fixed_loss()\n\tcriterion.cuda()\n\n\ttrain_dataset = Real('./data/SIDD_train/', 320, args.ps) + Syn('./data/Syn_train/', 100, args.ps)\n\ttrain_loader = torch.utils.data.DataLoader(\n\t\ttrain_dataset, batch_size=args.bs, shuffle=True, num_workers=8, pin_memory=True, drop_last=True)\n\n\tfor epoch in range(cur_epoch, args.epochs + 1):\n\t\tloss = train(train_loader, model, criterion, optimizer)\n\t\tscheduler.step()\n\n\t\ttorch.save({\n\t\t\t'epoch': epoch + 1,\n\t\t\t'state_dict': model.state_dict(),\n\t\t\t'optimizer' : optimizer.state_dict(),\n\t\t\t'scheduler' : scheduler.state_dict()}, \n\t\t\tos.path.join(save_dir, 'checkpoint.pth.tar'))\n\n\t\tprint('Epoch [{0}]\\t'\n\t\t\t'lr: {lr:.6f}\\t'\n\t\t\t'Loss: {loss:.5f}'\n\t\t\t.format(\n\t\t\tepoch,\n\t\t\tlr=optimizer.param_groups[-1]['lr'],\n\t\t\tloss=loss))\n" }, { "path": "utils/__init__.py", "content": "from .common import AverageMeter, ListAverageMeter, read_img, hwc_to_chw, chw_to_hwc" }, { "path": "utils/common.py", "content": "import numpy as np\nimport cv2\n\n\nclass AverageMeter(object):\n\tdef __init__(self):\n\t\tself.reset()\n\n\tdef reset(self):\n\t\tself.val = 0\n\t\tself.avg = 0\n\t\tself.sum = 0\n\t\tself.count = 0\n\n\tdef update(self, val, n=1):\n\t\tself.val = val\n\t\tself.sum += val * n\n\t\tself.count += n\n\t\tself.avg = self.sum / self.count\n\n\nclass ListAverageMeter(object):\n\t\"\"\"Computes and stores the average and current values of a list\"\"\"\n\tdef __init__(self):\n\t\tself.len = 10000 # set up the maximum length\n\t\tself.reset()\n\n\tdef reset(self):\n\t\tself.val = [0] * self.len\n\t\tself.avg = [0] * self.len\n\t\tself.sum = [0] * self.len\n\t\tself.count = 0\n\n\tdef set_len(self, n):\n\t\tself.len = n\n\t\tself.reset()\n\n\tdef update(self, vals, n=1):\n\t\tassert len(vals) == self.len, 'length of vals not equal to self.len'\n\t\tself.val = vals\n\t\tfor i in range(self.len):\n\t\t\tself.sum[i] += self.val[i] * n\n\t\tself.count += n\n\t\tfor i in range(self.len):\n\t\t\tself.avg[i] = self.sum[i] / self.count\n\t\t\t\n\ndef read_img(filename):\n\timg = cv2.imread(filename)\n\timg = img[:,:,::-1] / 255.0\n\t\t\n\timg = np.array(img).astype('float32')\n\n\treturn img\n\n\ndef hwc_to_chw(img):\n\treturn np.transpose(img, axes=[2, 0, 1]).astype('float32')\n\n\ndef chw_to_hwc(img):\n\treturn np.transpose(img, axes=[1, 2, 0]).astype('float32')\n" }, { "path": "utils/syn/ISP_implement.py", "content": "import random\nimport numpy as np\nimport cv2\nimport os\nimport json\nimport scipy.io\nimport math\nimport skimage\n\nfrom modules import demosaicing_CFA_Bayer_Malvar2004, CRF_Map_Cython, ICRF_Map_Cython\n\n\nclass ISP:\n def __init__(self, curve_path='./'):\n filename = os.path.join(curve_path, 'metadata/201_CRF_data.mat')\n CRFs = scipy.io.loadmat(filename)\n self.I = CRFs['I']\n self.B = CRFs['B']\n filename = os.path.join(curve_path, 'metadata/dorfCurvesInv.mat')\n inverseCRFs = scipy.io.loadmat(filename)\n self.I_inv = inverseCRFs['invI']\n self.B_inv = inverseCRFs['invB']\n filename = os.path.join(curve_path, 'metadata/cameras.json')\n with open(filename, 'r') as load_f:\n self.cameras = json.load(load_f)\n\n def ICRF_Map(self, img):\n invI_temp = self.I_inv[self.icrf_index, :]\n invB_temp = self.B_inv[self.icrf_index, :]\n out = ICRF_Map_Cython(img.astype(np.double), invI_temp.astype(np.double), invB_temp.astype(np.double))\n return out\n\n def CRF_Map(self, img):\n I_temp = self.I[self.icrf_index, :] # shape: (1024, 1)\n B_temp = self.B[self.icrf_index, :] # shape: (1024, 1)\n out = CRF_Map_Cython(img.astype(np.double), I_temp.astype(np.double), B_temp.astype(np.double))\n return out\n\n def RGB2XYZ(self, img):\n xyz = skimage.color.rgb2xyz(img)\n return xyz\n\n def XYZ2RGB(self, img):\n rgb = skimage.color.xyz2rgb(img)\n return rgb\n\n def XYZ2CAM(self, img):\n M_xyz2cam = np.reshape(self.M_xyz2cam, (3, 3))\n M_xyz2cam = M_xyz2cam / np.tile(np.sum(M_xyz2cam, axis=1), [3, 1]).T\n cam = self.apply_cmatrix(img, M_xyz2cam)\n cam = np.clip(cam, 0, 1)\n return cam\n\n def CAM2XYZ(self, img):\n M_xyz2cam = np.reshape(self.M_xyz2cam, (3, 3))\n M_xyz2cam = M_xyz2cam / np.tile(np.sum(M_xyz2cam, axis=1), [3, 1]).T\n M_cam2xyz = np.linalg.inv(M_xyz2cam)\n xyz = self.apply_cmatrix(img, M_cam2xyz)\n xyz = np.clip(xyz, 0, 1)\n return xyz\n\n def apply_cmatrix(self, img, matrix):\n r = (matrix[0, 0] * img[:, :, 0] + matrix[0, 1] * img[:, :, 1]\n + matrix[0, 2] * img[:, :, 2])\n g = (matrix[1, 0] * img[:, :, 0] + matrix[1, 1] * img[:, :, 1]\n + matrix[1, 2] * img[:, :, 2])\n b = (matrix[2, 0] * img[:, :, 0] + matrix[2, 1] * img[:, :, 1]\n + matrix[2, 2] * img[:, :, 2])\n r = np.expand_dims(r, axis=2)\n g = np.expand_dims(g, axis=2)\n b = np.expand_dims(b, axis=2)\n results = np.concatenate((r, g, b), axis=2)\n return results\n\n def mosaic_bayer(self, rgb):\n # analysis pattern\n num = np.zeros(4, dtype=int)\n # the image store in OpenCV using BGR\n temp = list(self.find(self.pattern, 'R'))\n num[temp] = 0\n temp = list(self.find(self.pattern, 'G'))\n num[temp] = 1\n temp = list(self.find(self.pattern, 'B'))\n num[temp] = 2\n\n mosaic_img = np.zeros((rgb.shape[0], rgb.shape[1]), dtype=rgb.dtype)\n mosaic_img[0::2, 0::2] = rgb[0::2, 0::2, num[0]]\n mosaic_img[0::2, 1::2] = rgb[0::2, 1::2, num[1]]\n mosaic_img[1::2, 0::2] = rgb[1::2, 0::2, num[2]]\n mosaic_img[1::2, 1::2] = rgb[1::2, 1::2, num[3]]\n return mosaic_img\n\n def WB_Mask(self, img, fr_now, fb_now):\n wb_mask = np.ones(img.shape)\n if self.pattern == 'RGGB':\n wb_mask[0::2, 0::2] = fr_now\n wb_mask[1::2, 1::2] = fb_now\n elif self.pattern == 'BGGR':\n wb_mask[1::2, 1::2] = fr_now\n wb_mask[0::2, 0::2] = fb_now\n elif self.pattern == 'GRBG':\n wb_mask[0::2, 1::2] = fr_now\n wb_mask[1::2, 0::2] = fb_now\n elif self.pattern == 'GBRG':\n wb_mask[1::2, 0::2] = fr_now\n wb_mask[0::2, 1::2] = fb_now\n return wb_mask\n\n\n def find(self, str, ch):\n for i, ltr in enumerate(str):\n if ltr == ch:\n yield i\n\n def Demosaic(self, bayer):\n results = demosaicing_CFA_Bayer_Malvar2004(bayer, self.pattern)\n results = np.clip(results, 0, 1)\n return results\n \n def add_PG_noise(self, img):\n min_log = np.log([0.0001])\n max_log_s = np.log([0.01])\n\n log_sigma_s = min_log + np.random.rand(1) * (max_log_s - min_log)\n sigma_s = np.exp(log_sigma_s)\n\n line_c = 2.2 * log_sigma_s + 1.2\n offset_c = np.random.normal(0.0, 0.26)\n log_sigma_c = line_c + offset_c\n sigma_c = np.exp(log_sigma_c)\n\n # add noise\n sigma_total = np.sqrt(sigma_s * img + sigma_c)\n\n noisy_img = img + \\\n sigma_total * np.random.randn(img.shape[0], img.shape[1])\n return noisy_img, sigma_s, sigma_c\n\n def noise_generate_srgb(self, img, configs='DND'):\n # -------------------------- CAMERA SETTING --------------------------\n cameras = self.cameras[configs]\n camera = cameras[random.randint(0, len(cameras)-1)]\n\n self.icrf_index = random.randint(0, 200)\n\n try:\n self.pattern = camera['bayertype']\n except:\n self.pattern = random.choice(['GRBG', 'RGGB', 'GBRG', 'BGGR'])\n\n try:\n ColorMatrix1 = camera['ColorMatrix1']\n ColorMatrix2 = camera['ColorMatrix2']\n alpha = np.random.random_sample([1])\n self.M_xyz2cam = alpha * ColorMatrix1 + (1 - alpha) * ColorMatrix2\n except:\n cam_index = np.random.random((1, 4))\n cam_index = cam_index / np.sum(cam_index)\n self.M_xyz2cam = ([1.0234,-0.2969,-0.2266,-0.5625,1.6328,-0.0469,-0.0703,0.2188,0.6406] * cam_index[0, 0] + \\\n [0.4913,-0.0541,-0.0202,-0.613,1.3513,0.2906,-0.1564,0.2151,0.7183] * cam_index[0, 1] + \\\n [0.838,-0.263,-0.0639,-0.2887,1.0725,0.2496,-0.0627,0.1427,0.5438] * cam_index[0, 2] + \\\n [0.6596,-0.2079,-0.0562,-0.4782,1.3016,0.1933,-0.097,0.1581,0.5181] * cam_index[0, 3])\n\n try:\n min_offset = -0.05\n max_offset = 0.05\n AsShotNeutral = camera['AsShotNeutral']\n self.fr_now = AsShotNeutral[0] + random.uniform(min_offset, max_offset)\n self.fb_now = AsShotNeutral[2] + random.uniform(min_offset, max_offset)\n except:\n min_fc = 0.75\n max_fc = 1\n self.fr_now = random.uniform(min_fc, max_fc)\n self.fb_now = random.uniform(min_fc, max_fc)\n \n try:\n blacklevel = camera['blacklevel']\n whitelevel = camera['whitelevel']\n except:\n blacklevel = 254\n whitelevel = 4094\n \n # -------------------------- INVERSE ISP PROCESS --------------------------\n img_rgb = img\n # Step 1 : inverse tone mapping\n img_L = self.ICRF_Map(img_rgb)\n # Step 2 : from RGB to XYZ\n img_XYZ = self.RGB2XYZ(img_L)\n # Step 3: from XYZ to Cam\n img_Cam = self.XYZ2CAM(img_XYZ)\n # Step 4: Mosaic\n img_mosaic = self.mosaic_bayer(img_Cam)\n # Step 5: inverse White Balance\n wb_mask = self.WB_Mask(img_mosaic, self.fr_now, self.fb_now)\n img_mosaic = img_mosaic * wb_mask\n img_mosaic_gt = img_mosaic\n\n # -------------------------- POISSON-GAUSSIAN NOISE ON RAW --------------------------\n img_mosaic_noise, sigma_s, sigma_c = self.add_PG_noise(img_mosaic)\n\n # -------------------------- QUANTIZATION NOISE AND CLIPPING EFFECT ON RAW --------------------------\n upper_bound = math.pow(2, math.ceil(math.log(whitelevel + 1, 2))) - 1\n img_mosaic_noise = np.clip(np.floor(img_mosaic_noise * (whitelevel - blacklevel) + blacklevel), 0, upper_bound)\n img_mosaic_noise = (img_mosaic_noise - blacklevel) / (whitelevel - blacklevel)\n img_mosaic_gt = np.clip(np.floor(img_mosaic_gt * (whitelevel - blacklevel) + blacklevel), 0, upper_bound)\n img_mosaic_gt = (img_mosaic_gt - blacklevel) / (whitelevel - blacklevel)\n\n # -------------------------- ISP PROCESS --------------------------\n # Step 5 : White Balance\n wb_mask = self.WB_Mask(img_mosaic_noise, 1/self.fr_now, 1/self.fb_now)\n img_mosaic_noise = img_mosaic_noise * wb_mask\n img_mosaic_noise = np.clip(img_mosaic_noise, 0, 1)\n img_mosaic_gt = img_mosaic_gt * wb_mask\n img_mosaic_gt = np.clip(img_mosaic_gt, 0, 1)\n # Step 4 : Demosaic\n img_demosaic = self.Demosaic(img_mosaic_noise)\n img_demosaic_gt = self.Demosaic(img_mosaic_gt)\n # Step 3 : from Cam to XYZ\n img_IXYZ = self.CAM2XYZ(img_demosaic)\n img_IXYZ_gt = self.CAM2XYZ(img_demosaic_gt)\n # Step 2 : frome XYZ to RGB\n img_IL = self.XYZ2RGB(img_IXYZ)\n img_IL_gt = self.XYZ2RGB(img_IXYZ_gt)\n # Step 1 : tone mapping\n img_Irgb = self.CRF_Map(img_IL)\n img_Irgb_gt = self.CRF_Map(img_IL_gt)\n\n # -------------------------- QUANTIZATION NOISE AND CLIPPING EFFECT ON RGB --------------------------\n noise = np.clip(img_Irgb, 0, 1) - np.clip(img_Irgb_gt, 0, 1)\n img_Irgb_gt = np.clip(img_rgb, 0, 1)\n img_Irgb = np.clip((img_rgb + noise), 0, 1)\n\n sigma_total = np.sqrt(sigma_s * img + sigma_c) # noise level map\n\n return np.uint8(np.round(img_Irgb_gt*255)), np.uint8(np.round(img_Irgb*255)), sigma_total\n" }, { "path": "utils/syn/generate_dataset.py", "content": "import os\nimport random, math\nimport torch\nimport numpy as np\nimport glob\nimport cv2\nfrom tqdm import tqdm\nfrom skimage import io\n\nfrom ISP_implement import ISP\n\n\nif __name__ == '__main__':\n\tisp = ISP()\n\n\tsource_dir = './source/'\n\ttarget_dir = './target/'\n\n\tif not os.path.isdir(target_dir):\n\t\tos.makedirs(target_dir)\n\n\tfns = glob.glob(os.path.join(source_dir, '*.png'))\n\n\tpatch_size = 256\n\n\tfor fn in tqdm(fns):\n\t\timg_rgb = cv2.imread(fn)[:, :, ::-1] / 255.0\n\n\t\tH = img_rgb.shape[0]\n\t\tW = img_rgb.shape[1]\n\n\t\tH_s = H // patch_size\n\t\tW_s = W // patch_size\n\n\t\tpatch_id = 0\n\n\t\tfor i in range(H_s):\n\t\t\tfor j in range(W_s):\n\t\n\t\t\t\tyy = i * patch_size\n\t\t\t\txx = j * patch_size\n\n\t\t\t\tpatch_img_rgb = img_rgb[yy:yy+patch_size, xx:xx+patch_size, :]\n\n\t\t\t\tgt, noise, sigma = isp.noise_generate_srgb(patch_img_rgb)\n\n\t\t\t\tsigma = np.uint8(np.round(np.clip(sigma * 15 , 0, 1) * 255))\t# store in uint8\n\n\t\t\t\tfilename = os.path.basename(fn)\n\t\t\t\tfoldername = filename.split('.')[0]\n\n\t\t\t\tout_folder = os.path.join(target_dir, foldername)\n\n\t\t\t\tif not os.path.isdir(out_folder):\n\t\t\t\t\tos.makedirs(out_folder)\n\n\t\t\t\tio.imsave(os.path.join(out_folder, 'GT_SRGB_%d_%d.png' % (i, j)), gt)\n\t\t\t\tio.imsave(os.path.join(out_folder, 'NOISY_SRGB_%d_%d.png' % (i, j)), noise)\n\t\t\t\tio.imsave(os.path.join(out_folder, 'SIGMA_SRGB_%d_%d.png' % (i, j)), sigma)\n" }, { "path": "utils/syn/metadata/cameras.json", "content": "{\n \"DND\": [\n {\n \"bayertype\": \"GBRG\",\n \"blacklevel\": 52.0,\n \"whitelevel\": 1023.0,\n \"ColorMatrix1\": [\n 0.8203,\n -0.2266,\n -0.125,\n -0.3203,\n 1.2656,\n 0.0391,\n -0.0391,\n 0.2266,\n 0.4531\n ],\n \"ColorMatrix2\": [\n 1.0234,\n -0.2969,\n -0.2266,\n -0.5625,\n 1.6328,\n -0.0469,\n -0.0703,\n 0.2188,\n 0.6406\n ],\n \"AsShotNeutral\": [\n 0.4922,\n 1.0078,\n 0.6016\n ]\n },\n {\n \"bayertype\": \"RGGB\",\n \"blacklevel\": 256.0,\n \"whitelevel\": 7680.0,\n \"ColorMatrix1\": [\n 0.7216,\n -0.2921,\n 0.035,\n -0.4204,\n 1.1461,\n 0.3143,\n -0.0767,\n 0.1485,\n 0.7418\n ],\n \"ColorMatrix2\": [\n 0.4913,\n -0.0541,\n -0.0202,\n -0.613,\n 1.3513,\n 0.2906,\n -0.1564,\n 0.2151,\n 0.7183\n ],\n \"AsShotNeutral\": [\n 0.419,\n 1.0,\n 0.6275\n ]\n },\n {\n \"bayertype\": \"RGGB\",\n \"blacklevel\": 254.0,\n \"whitelevel\": 4094.0,\n \"ColorMatrix1\": [\n 0.9033,\n -0.3597,\n 0.026,\n -0.2351,\n 0.97,\n 0.3111,\n -0.0181,\n 0.0807,\n 0.5838\n ],\n \"ColorMatrix2\": [\n 0.838,\n -0.263,\n -0.0639,\n -0.2887,\n 1.0725,\n 0.2496,\n -0.0627,\n 0.1427,\n 0.5438\n ],\n \"AsShotNeutral\": [\n 0.5246,\n 1.0,\n 0.5424\n ]\n },\n {\n \"bayertype\": \"GBRG\",\n \"blacklevel\": 254.0,\n \"whitelevel\": 4094.0,\n \"ColorMatrix1\": [\n 0.9033,\n -0.3597,\n 0.026,\n -0.2351,\n 0.97,\n 0.3111,\n -0.0181,\n 0.0807,\n 0.5838\n ],\n \"ColorMatrix2\": [\n 0.838,\n -0.263,\n -0.0639,\n -0.2887,\n 1.0725,\n 0.2496,\n -0.0627,\n 0.1427,\n 0.5438\n ],\n \"AsShotNeutral\": [\n 0.5246,\n 1.0,\n 0.5424\n ]\n },\n {\n \"bayertype\": \"RGGB\",\n \"blacklevel\": 400.0,\n \"whitelevel\": 7680.0,\n \"ColorMatrix1\": [\n 0.7366,\n -0.3213,\n 0.038,\n -0.3609,\n 1.1127,\n 0.2852,\n -0.0218,\n 0.0694,\n 0.5821\n ],\n \"ColorMatrix2\": [\n 0.6596,\n -0.2079,\n -0.0562,\n -0.4782,\n 1.3016,\n 0.1933,\n -0.097,\n 0.1581,\n 0.5181\n ],\n \"AsShotNeutral\": [\n 0.4044,\n 1.0,\n 0.5356\n ]\n },\n {\n \"bayertype\": \"GRBG\",\n \"blacklevel\": 400.0,\n \"whitelevel\": 7680.0,\n \"ColorMatrix1\": [\n 0.7366,\n -0.3213,\n 0.038,\n -0.3609,\n 1.1127,\n 0.2852,\n -0.0218,\n 0.0694,\n 0.5821\n ],\n \"ColorMatrix2\": [\n 0.6596,\n -0.2079,\n -0.0562,\n -0.4782,\n 1.3016,\n 0.1933,\n -0.097,\n 0.1581,\n 0.5181\n ],\n \"AsShotNeutral\": [\n 0.4044,\n 1.0,\n 0.5356\n ]\n }\n ],\n \"SIDD\": [\n {\n \"ColorMatrix1\": [\n 0.6640625,\n -0.0458984375,\n -0.1201171875,\n -0.54296875,\n 1.4384765625,\n 0.0712890625,\n -0.1767578125,\n 0.4052734375,\n 0.4755859375\n ],\n \"ColorMatrix2\": [\n 1.23828125,\n -0.4443359375,\n -0.2783203125,\n -0.4501953125,\n 1.4697265625,\n 0.0693359375,\n -0.08984375,\n 0.2919921875,\n 0.61328125\n ],\n \"AsShotNeutral\": [\n 0.56640625,\n 1.0,\n 0.4951171875\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.7265625,\n -0.1953125,\n -0.0859375,\n -0.5625,\n 1.3515625,\n 0.1640625,\n -0.2265625,\n 0.3046875,\n 0.53125\n ],\n \"ColorMatrix2\": [\n 1.0703125,\n -0.3125,\n -0.28125,\n -0.5625,\n 1.65625,\n -0.1171875,\n -0.0546875,\n 0.1875,\n 0.5859375\n ],\n \"AsShotNeutral\": [\n 0.4140625,\n 1.0,\n 0.5859375\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.6796875,\n -0.078125,\n -0.09375,\n -0.4609375,\n 1.296875,\n 0.1328125,\n -0.109375,\n 0.25,\n 0.5234375\n ],\n \"ColorMatrix2\": [\n 1.1875,\n -0.4140625,\n -0.25,\n -0.4609375,\n 1.5,\n 0.015625,\n -0.046875,\n 0.2109375,\n 0.59375\n ],\n \"AsShotNeutral\": [\n 0.484375,\n 1.0,\n 0.6328125\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.5859375,\n 0.0546875,\n -0.125,\n -0.6484375,\n 1.5546875,\n 0.0546875,\n -0.2421875,\n 0.5625,\n 0.390625\n ],\n \"ColorMatrix2\": [\n 1.15625,\n -0.2890625,\n -0.3203125,\n -0.53125,\n 1.5625,\n 0.0625,\n -0.078125,\n 0.28125,\n 0.5625\n ],\n \"AsShotNeutral\": [\n 0.4609375,\n 1.0,\n 0.6875\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.8353,\n -0.3171,\n -0.1289,\n -0.3878,\n 1.1893,\n 0.2237,\n -0.038,\n 0.1056,\n 0.6397\n ],\n \"ColorMatrix2\": [\n 0.7418,\n -0.2398,\n -0.061,\n -0.5006,\n 1.2972,\n 0.2248,\n -0.1074,\n 0.1419,\n 0.59\n ],\n \"AsShotNeutral\": [\n 0.406349,\n 1.0,\n 0.601468\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.7265625,\n -0.1953125,\n -0.0859375,\n -0.5625,\n 1.3515625,\n 0.1640625,\n -0.2265625,\n 0.3046875,\n 0.53125\n ],\n \"ColorMatrix2\": [\n 1.0703125,\n -0.3125,\n -0.28125,\n -0.5625,\n 1.65625,\n -0.1171875,\n -0.0546875,\n 0.1875,\n 0.5859375\n ],\n \"AsShotNeutral\": [\n 0.5703125,\n 1.0078125,\n 0.5078125\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.8353,\n -0.3171,\n -0.1289,\n -0.3878,\n 1.1893,\n 0.2237,\n -0.038,\n 0.1056,\n 0.6397\n ],\n \"ColorMatrix2\": [\n 0.7418,\n -0.2398,\n -0.061,\n -0.5006,\n 1.2972,\n 0.2248,\n -0.1074,\n 0.1419,\n 0.59\n ],\n \"AsShotNeutral\": [\n 0.407806,\n 1.0,\n 0.640901\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.5859375,\n 0.0546875,\n -0.125,\n -0.6484375,\n 1.5546875,\n 0.0546875,\n -0.2421875,\n 0.5625,\n 0.390625\n ],\n \"ColorMatrix2\": [\n 1.15625,\n -0.2890625,\n -0.3203125,\n -0.53125,\n 1.5625,\n 0.0625,\n -0.078125,\n 0.28125,\n 0.5625\n ],\n \"AsShotNeutral\": [\n 0.640625,\n 1.0,\n 0.515625\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.6796875,\n -0.078125,\n -0.09375,\n -0.4609375,\n 1.296875,\n 0.1328125,\n -0.109375,\n 0.25,\n 0.5234375\n ],\n \"ColorMatrix2\": [\n 1.1875,\n -0.4140625,\n -0.25,\n -0.4609375,\n 1.5,\n 0.015625,\n -0.046875,\n 0.2109375,\n 0.59375\n ],\n \"AsShotNeutral\": [\n 0.71875,\n 1.0,\n 0.4921875\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.6640625,\n -0.0458984375,\n -0.1201171875,\n -0.54296875,\n 1.4384765625,\n 0.0712890625,\n -0.1767578125,\n 0.4052734375,\n 0.4755859375\n ],\n \"ColorMatrix2\": [\n 1.23828125,\n -0.4443359375,\n -0.2783203125,\n -0.4501953125,\n 1.4697265625,\n 0.0693359375,\n -0.08984375,\n 0.2919921875,\n 0.61328125\n ],\n \"AsShotNeutral\": [\n 0.5478515625,\n 1.0,\n 0.576171875\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.6796875,\n -0.078125,\n -0.09375,\n -0.4609375,\n 1.296875,\n 0.1328125,\n -0.109375,\n 0.25,\n 0.5234375\n ],\n \"ColorMatrix2\": [\n 1.1875,\n -0.4140625,\n -0.25,\n -0.4609375,\n 1.5,\n 0.015625,\n -0.046875,\n 0.2109375,\n 0.59375\n ],\n \"AsShotNeutral\": [\n 0.546875,\n 1.0,\n 0.625\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.5859375,\n 0.0546875,\n -0.125,\n -0.6484375,\n 1.5546875,\n 0.0546875,\n -0.2421875,\n 0.5625,\n 0.390625\n ],\n \"ColorMatrix2\": [\n 1.15625,\n -0.2890625,\n -0.3203125,\n -0.53125,\n 1.5625,\n 0.0625,\n -0.078125,\n 0.28125,\n 0.5625\n ],\n \"AsShotNeutral\": [\n 0.515625,\n 1.0,\n 0.65625\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.6640625,\n -0.0458984375,\n -0.1201171875,\n -0.54296875,\n 1.4384765625,\n 0.0712890625,\n -0.1767578125,\n 0.4052734375,\n 0.4755859375\n ],\n \"ColorMatrix2\": [\n 1.23828125,\n -0.4443359375,\n -0.2783203125,\n -0.4501953125,\n 1.4697265625,\n 0.0693359375,\n -0.08984375,\n 0.2919921875,\n 0.61328125\n ],\n \"AsShotNeutral\": [\n 0.583984375,\n 1.0,\n 0.5869140625\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.7265625,\n -0.1953125,\n -0.0859375,\n -0.5625,\n 1.3515625,\n 0.1640625,\n -0.2265625,\n 0.3046875,\n 0.53125\n ],\n \"ColorMatrix2\": [\n 1.0703125,\n -0.3125,\n -0.28125,\n -0.5625,\n 1.65625,\n -0.1171875,\n -0.0546875,\n 0.1875,\n 0.5859375\n ],\n \"AsShotNeutral\": [\n 0.5546875,\n 0.9921875,\n 0.5078125\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.8353,\n -0.3171,\n -0.1289,\n -0.3878,\n 1.1893,\n 0.2237,\n -0.038,\n 0.1056,\n 0.6397\n ],\n \"ColorMatrix2\": [\n 0.7418,\n -0.2398,\n -0.061,\n -0.5006,\n 1.2972,\n 0.2248,\n -0.1074,\n 0.1419,\n 0.59\n ],\n \"AsShotNeutral\": [\n 0.5752,\n 1.0,\n 0.407076\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.6796875,\n -0.078125,\n -0.09375,\n -0.4609375,\n 1.296875,\n 0.1328125,\n -0.109375,\n 0.25,\n 0.5234375\n ],\n \"ColorMatrix2\": [\n 1.1875,\n -0.4140625,\n -0.25,\n -0.4609375,\n 1.5,\n 0.015625,\n -0.046875,\n 0.2109375,\n 0.59375\n ],\n \"AsShotNeutral\": [\n 0.796875,\n 0.9921875,\n 0.390625\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.5859375,\n 0.0546875,\n -0.125,\n -0.6484375,\n 1.5546875,\n 0.0546875,\n -0.2421875,\n 0.5625,\n 0.390625\n ],\n \"ColorMatrix2\": [\n 1.15625,\n -0.2890625,\n -0.3203125,\n -0.53125,\n 1.5625,\n 0.0625,\n -0.078125,\n 0.28125,\n 0.5625\n ],\n \"AsShotNeutral\": [\n 0.71875,\n 1.0,\n 0.484375\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.6640625,\n -0.0458984375,\n -0.1201171875,\n -0.54296875,\n 1.4384765625,\n 0.0712890625,\n -0.1767578125,\n 0.4052734375,\n 0.4755859375\n ],\n \"ColorMatrix2\": [\n 1.23828125,\n -0.4443359375,\n -0.2783203125,\n -0.4501953125,\n 1.4697265625,\n 0.0693359375,\n -0.08984375,\n 0.2919921875,\n 0.61328125\n ],\n \"AsShotNeutral\": [\n 0.751953125,\n 1.0,\n 0.59375\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.7265625,\n -0.1953125,\n -0.0859375,\n -0.5625,\n 1.3515625,\n 0.1640625,\n -0.2265625,\n 0.3046875,\n 0.53125\n ],\n \"ColorMatrix2\": [\n 1.0703125,\n -0.3125,\n -0.28125,\n -0.5625,\n 1.65625,\n -0.1171875,\n -0.0546875,\n 0.1875,\n 0.5859375\n ],\n \"AsShotNeutral\": [\n 0.65625,\n 0.9921875,\n 0.4296875\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.8353,\n -0.3171,\n -0.1289,\n -0.3878,\n 1.1893,\n 0.2237,\n -0.038,\n 0.1056,\n 0.6397\n ],\n \"ColorMatrix2\": [\n 0.7418,\n -0.2398,\n -0.061,\n -0.5006,\n 1.2972,\n 0.2248,\n -0.1074,\n 0.1419,\n 0.59\n ],\n \"AsShotNeutral\": [\n 0.590712,\n 1.0,\n 0.453448\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.6796875,\n -0.078125,\n -0.09375,\n -0.4609375,\n 1.296875,\n 0.1328125,\n -0.109375,\n 0.25,\n 0.5234375\n ],\n \"ColorMatrix2\": [\n 1.1875,\n -0.4140625,\n -0.25,\n -0.4609375,\n 1.5,\n 0.015625,\n -0.046875,\n 0.2109375,\n 0.59375\n ],\n \"AsShotNeutral\": [\n 0.609375,\n 1.0,\n 0.5\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.5859375,\n 0.0546875,\n -0.125,\n -0.6484375,\n 1.5546875,\n 0.0546875,\n -0.2421875,\n 0.5625,\n 0.390625\n ],\n \"ColorMatrix2\": [\n 1.15625,\n -0.2890625,\n -0.3203125,\n -0.53125,\n 1.5625,\n 0.0625,\n -0.078125,\n 0.28125,\n 0.5625\n ],\n \"AsShotNeutral\": [\n 0.5625,\n 1.0,\n 0.515625\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.6640625,\n -0.0458984375,\n -0.1201171875,\n -0.54296875,\n 1.4384765625,\n 0.0712890625,\n -0.1767578125,\n 0.4052734375,\n 0.4755859375\n ],\n \"ColorMatrix2\": [\n 1.23828125,\n -0.4443359375,\n -0.2783203125,\n -0.4501953125,\n 1.4697265625,\n 0.0693359375,\n -0.08984375,\n 0.2919921875,\n 0.61328125\n ],\n \"AsShotNeutral\": [\n 0.4951171875,\n 1.0,\n 0.576171875\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.7265625,\n -0.1953125,\n -0.0859375,\n -0.5625,\n 1.3515625,\n 0.1640625,\n -0.2265625,\n 0.3046875,\n 0.53125\n ],\n \"ColorMatrix2\": [\n 1.0703125,\n -0.3125,\n -0.28125,\n -0.5625,\n 1.65625,\n -0.1171875,\n -0.0546875,\n 0.1875,\n 0.5859375\n ],\n \"AsShotNeutral\": [\n 0.4453125,\n 0.9921875,\n 0.53125\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.8353,\n -0.3171,\n -0.1289,\n -0.3878,\n 1.1893,\n 0.2237,\n -0.038,\n 0.1056,\n 0.6397\n ],\n \"ColorMatrix2\": [\n 0.7418,\n -0.2398,\n -0.061,\n -0.5006,\n 1.2972,\n 0.2248,\n -0.1074,\n 0.1419,\n 0.59\n ],\n \"AsShotNeutral\": [\n 0.343221,\n 1.0,\n 0.593365\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.6796875,\n -0.078125,\n -0.09375,\n -0.4609375,\n 1.296875,\n 0.1328125,\n -0.109375,\n 0.25,\n 0.5234375\n ],\n \"ColorMatrix2\": [\n 1.1875,\n -0.4140625,\n -0.25,\n -0.4609375,\n 1.5,\n 0.015625,\n -0.046875,\n 0.2109375,\n 0.59375\n ],\n \"AsShotNeutral\": [\n 0.578125,\n 1.0,\n 0.53125\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.5859375,\n 0.0546875,\n -0.125,\n -0.6484375,\n 1.5546875,\n 0.0546875,\n -0.2421875,\n 0.5625,\n 0.390625\n ],\n \"ColorMatrix2\": [\n 1.15625,\n -0.2890625,\n -0.3203125,\n -0.53125,\n 1.5625,\n 0.0625,\n -0.078125,\n 0.28125,\n 0.5625\n ],\n \"AsShotNeutral\": [\n 0.5859375,\n 1.0,\n 0.515625\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.6640625,\n -0.0458984375,\n -0.1201171875,\n -0.54296875,\n 1.4384765625,\n 0.0712890625,\n -0.1767578125,\n 0.4052734375,\n 0.4755859375\n ],\n \"ColorMatrix2\": [\n 1.23828125,\n -0.4443359375,\n -0.2783203125,\n -0.4501953125,\n 1.4697265625,\n 0.0693359375,\n -0.08984375,\n 0.2919921875,\n 0.61328125\n ],\n \"AsShotNeutral\": [\n 0.5234375,\n 1.0,\n 0.5947265625\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.7265625,\n -0.1953125,\n -0.0859375,\n -0.5625,\n 1.3515625,\n 0.1640625,\n -0.2265625,\n 0.3046875,\n 0.53125\n ],\n \"ColorMatrix2\": [\n 1.0703125,\n -0.3125,\n -0.28125,\n -0.5625,\n 1.65625,\n -0.1171875,\n -0.0546875,\n 0.1875,\n 0.5859375\n ],\n \"AsShotNeutral\": [\n 0.4140625,\n 0.9921875,\n 0.5703125\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.8353,\n -0.3171,\n -0.1289,\n -0.3878,\n 1.1893,\n 0.2237,\n -0.038,\n 0.1056,\n 0.6397\n ],\n \"ColorMatrix2\": [\n 0.7418,\n -0.2398,\n -0.061,\n -0.5006,\n 1.2972,\n 0.2248,\n -0.1074,\n 0.1419,\n 0.59\n ],\n \"AsShotNeutral\": [\n 0.383592,\n 1.0,\n 0.555993\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.6796875,\n -0.078125,\n -0.09375,\n -0.4609375,\n 1.296875,\n 0.1328125,\n -0.109375,\n 0.25,\n 0.5234375\n ],\n \"ColorMatrix2\": [\n 1.1875,\n -0.4140625,\n -0.25,\n -0.4609375,\n 1.5,\n 0.015625,\n -0.046875,\n 0.2109375,\n 0.59375\n ],\n \"AsShotNeutral\": [\n 0.625,\n 1.0,\n 0.609375\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.5859375,\n 0.0546875,\n -0.125,\n -0.6484375,\n 1.5546875,\n 0.0546875,\n -0.2421875,\n 0.5625,\n 0.390625\n ],\n \"ColorMatrix2\": [\n 1.15625,\n -0.2890625,\n -0.3203125,\n -0.53125,\n 1.5625,\n 0.0625,\n -0.078125,\n 0.28125,\n 0.5625\n ],\n \"AsShotNeutral\": [\n 0.578125,\n 1.0,\n 0.609375\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.6640625,\n -0.0458984375,\n -0.1201171875,\n -0.54296875,\n 1.4384765625,\n 0.0712890625,\n -0.1767578125,\n 0.4052734375,\n 0.4755859375\n ],\n \"ColorMatrix2\": [\n 1.23828125,\n -0.4443359375,\n -0.2783203125,\n -0.4501953125,\n 1.4697265625,\n 0.0693359375,\n -0.08984375,\n 0.2919921875,\n 0.61328125\n ],\n \"AsShotNeutral\": [\n 0.5556640625,\n 1.0,\n 0.626953125\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.7265625,\n -0.1953125,\n -0.0859375,\n -0.5625,\n 1.3515625,\n 0.1640625,\n -0.2265625,\n 0.3046875,\n 0.53125\n ],\n \"ColorMatrix2\": [\n 1.0703125,\n -0.3125,\n -0.28125,\n -0.5625,\n 1.65625,\n -0.1171875,\n -0.0546875,\n 0.1875,\n 0.5859375\n ],\n \"AsShotNeutral\": [\n 0.546875,\n 0.9921875,\n 0.4375\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.8353,\n -0.3171,\n -0.1289,\n -0.3878,\n 1.1893,\n 0.2237,\n -0.038,\n 0.1056,\n 0.6397\n ],\n \"ColorMatrix2\": [\n 0.7418,\n -0.2398,\n -0.061,\n -0.5006,\n 1.2972,\n 0.2248,\n -0.1074,\n 0.1419,\n 0.59\n ],\n \"AsShotNeutral\": [\n 0.524523,\n 1.0,\n 0.480638\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.6796875,\n -0.078125,\n -0.09375,\n -0.4609375,\n 1.296875,\n 0.1328125,\n -0.109375,\n 0.25,\n 0.5234375\n ],\n \"ColorMatrix2\": [\n 1.1875,\n -0.4140625,\n -0.25,\n -0.4609375,\n 1.5,\n 0.015625,\n -0.046875,\n 0.2109375,\n 0.59375\n ],\n \"AsShotNeutral\": [\n 0.71875,\n 1.0,\n 0.5234375\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.5859375,\n 0.0546875,\n -0.125,\n -0.6484375,\n 1.5546875,\n 0.0546875,\n -0.2421875,\n 0.5625,\n 0.390625\n ],\n \"ColorMatrix2\": [\n 1.15625,\n -0.2890625,\n -0.3203125,\n -0.53125,\n 1.5625,\n 0.0625,\n -0.078125,\n 0.28125,\n 0.5625\n ],\n \"AsShotNeutral\": [\n 0.59375,\n 1.0,\n 0.5546875\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.6640625,\n -0.0458984375,\n -0.1201171875,\n -0.54296875,\n 1.4384765625,\n 0.0712890625,\n -0.1767578125,\n 0.4052734375,\n 0.4755859375\n ],\n \"ColorMatrix2\": [\n 1.23828125,\n -0.4443359375,\n -0.2783203125,\n -0.4501953125,\n 1.4697265625,\n 0.0693359375,\n -0.08984375,\n 0.2919921875,\n 0.61328125\n ],\n \"AsShotNeutral\": [\n 0.6005859375,\n 1.0,\n 0.6416015625\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.7265625,\n -0.1953125,\n -0.0859375,\n -0.5625,\n 1.3515625,\n 0.1640625,\n -0.2265625,\n 0.3046875,\n 0.53125\n ],\n \"ColorMatrix2\": [\n 1.0703125,\n -0.3125,\n -0.28125,\n -0.5625,\n 1.65625,\n -0.1171875,\n -0.0546875,\n 0.1875,\n 0.5859375\n ],\n \"AsShotNeutral\": [\n 0.546875,\n 0.9921875,\n 0.4375\n ]\n },\n {\n \"ColorMatrix1\": [\n 0.8353,\n -0.3171,\n -0.1289,\n -0.3878,\n 1.1893,\n 0.2237,\n -0.038,\n 0.1056,\n 0.6397\n ],\n \"ColorMatrix2\": [\n 0.7418,\n -0.2398,\n -0.061,\n -0.5006,\n 1.2972,\n 0.2248,\n -0.1074,\n 0.1419,\n 0.59\n ],\n \"AsShotNeutral\": [\n 0.630736,\n 1.0,\n 0.419586\n ]\n }\n ]\n}" }, { "path": "utils/syn/modules/Demosaicing_malvar2004.py", "content": "# -*- coding: utf-8 -*-\n\"\"\"\nMalvar (2004) Bayer CFA Demosaicing\n===================================\n\n*Bayer* CFA (Colour Filter Array) *Malvar (2004)* demosaicing.\n\nReferences\n----------\n- :cite:`Malvar2004a` : Malvar, H. S., He, L.-W., Cutler, R., & Way, O. M.\n (2004). High-Quality Linear Interpolation for Demosaicing of\n Bayer-Patterned Color Images. In International Conference of Acoustic,\n Speech and Signal Processing (pp. 5-8). Institute of Electrical and\n Electronics Engineers, Inc. Retrieved from\n http://research.microsoft.com/apps/pubs/default.aspx?id=102068\n https://colour-demosaicing.readthedocs.io/en/develop/_modules/colour_demosaicing/bayer/demosaicing/malvar2004.html\n\"\"\"\n\nfrom __future__ import division, unicode_literals\n\nimport numpy as np\nfrom scipy.ndimage.filters import convolve\n\n#from colour.utilities import tstack\nimport cv2\n\nfrom .masks import masks_CFA_Bayer\n\n__author__ = 'Colour Developers'\n__copyright__ = 'Copyright (C) 2015-2018 - Colour Developers'\n__license__ = 'New BSD License - http://opensource.org/licenses/BSD-3-Clause'\n__maintainer__ = 'Colour Developers'\n__email__ = 'colour-science@googlegroups.com'\n__status__ = 'Production'\n\n__all__ = ['demosaicing_CFA_Bayer_Malvar2004']\n\n\ndef demosaicing_CFA_Bayer_Malvar2004(CFA, pattern='RGGB'):\n \"\"\"\n Returns the demosaiced *RGB* colourspace array from given *Bayer* CFA using\n *Malvar (2004)* demosaicing algorithm.\n\n Parameters\n ----------\n CFA : array_like\n *Bayer* CFA.\n pattern : unicode, optional\n **{'RGGB', 'BGGR', 'GRBG', 'GBRG'}**,\n Arrangement of the colour filters on the pixel array.\n\n Returns\n -------\n ndarray\n *RGB* colourspace array.\n\n Notes\n -----\n - The definition output is not clipped in range [0, 1] : this allows for\n direct HDRI / radiance image generation on *Bayer* CFA data and post\n demosaicing of the high dynamic range data as showcased in this\n `Jupyter Notebook `_.\n\n References\n ----------\n - :cite:`Malvar2004a`\n\n Examples\n --------\n >>> CFA = np.array(\n ... [[0.30980393, 0.36078432, 0.30588236, 0.3764706],\n ... [0.35686275, 0.39607844, 0.36078432, 0.40000001]])\n >>> demosaicing_CFA_Bayer_Malvar2004(CFA)\n array([[[ 0.30980393, 0.31666668, 0.32941177],\n [ 0.33039216, 0.36078432, 0.38112746],\n [ 0.30588236, 0.32794118, 0.34877452],\n [ 0.36274511, 0.3764706 , 0.38480393]],\n \n [[ 0.34828432, 0.35686275, 0.36568628],\n [ 0.35318628, 0.38186275, 0.39607844],\n [ 0.3379902 , 0.36078432, 0.3754902 ],\n [ 0.37769609, 0.39558825, 0.40000001]]])\n >>> CFA = np.array(\n ... [[0.3764706, 0.360784320, 0.40784314, 0.3764706],\n ... [0.35686275, 0.30980393, 0.36078432, 0.29803923]])\n >>> demosaicing_CFA_Bayer_Malvar2004(CFA, 'BGGR')\n array([[[ 0.35539217, 0.37058825, 0.3764706 ],\n [ 0.34264707, 0.36078432, 0.37450981],\n [ 0.36568628, 0.39607844, 0.40784314],\n [ 0.36568629, 0.3764706 , 0.3882353 ]],\n \n [[ 0.34411765, 0.35686275, 0.36200981],\n [ 0.30980393, 0.32990197, 0.34975491],\n [ 0.33039216, 0.36078432, 0.38063726],\n [ 0.29803923, 0.30441178, 0.31740197]]])\n \"\"\"\n\n CFA = np.asarray(CFA)\n R_m, G_m, B_m = masks_CFA_Bayer(CFA.shape, pattern)\n\n GR_GB = np.asarray(\n [[0, 0, -1, 0, 0],\n [0, 0, 2, 0, 0],\n [-1, 2, 4, 2, -1],\n [0, 0, 2, 0, 0],\n [0, 0, -1, 0, 0]]) / 8 # yapf: disable\n\n Rg_RB_Bg_BR = np.asarray(\n [[0, 0, 0.5, 0, 0],\n [0, -1, 0, -1, 0],\n [-1, 4, 5, 4, - 1],\n [0, -1, 0, -1, 0],\n [0, 0, 0.5, 0, 0]]) / 8 # yapf: disable\n\n Rg_BR_Bg_RB = np.transpose(Rg_RB_Bg_BR)\n\n Rb_BB_Br_RR = np.asarray(\n [[0, 0, -1.5, 0, 0],\n [0, 2, 0, 2, 0],\n [-1.5, 0, 6, 0, -1.5],\n [0, 2, 0, 2, 0],\n [0, 0, -1.5, 0, 0]]) / 8 # yapf: disable\n\n R = CFA * R_m\n G = CFA * G_m\n B = CFA * B_m\n\n del G_m\n\n G = np.where(np.logical_or(R_m == 1, B_m == 1), convolve(CFA, GR_GB), G)\n\n RBg_RBBR = convolve(CFA, Rg_RB_Bg_BR)\n RBg_BRRB = convolve(CFA, Rg_BR_Bg_RB)\n RBgr_BBRR = convolve(CFA, Rb_BB_Br_RR)\n\n del GR_GB, Rg_RB_Bg_BR, Rg_BR_Bg_RB, Rb_BB_Br_RR\n\n # Red rows.\n R_r = np.transpose(np.any(R_m == 1, axis=1)[np.newaxis]) * np.ones(R.shape)\n # Red columns.\n R_c = np.any(R_m == 1, axis=0)[np.newaxis] * np.ones(R.shape)\n # Blue rows.\n B_r = np.transpose(np.any(B_m == 1, axis=1)[np.newaxis]) * np.ones(B.shape)\n # Blue columns\n B_c = np.any(B_m == 1, axis=0)[np.newaxis] * np.ones(B.shape)\n\n del R_m, B_m\n\n R = np.where(np.logical_and(R_r == 1, B_c == 1), RBg_RBBR, R)\n R = np.where(np.logical_and(B_r == 1, R_c == 1), RBg_BRRB, R)\n\n B = np.where(np.logical_and(B_r == 1, R_c == 1), RBg_RBBR, B)\n B = np.where(np.logical_and(R_r == 1, B_c == 1), RBg_BRRB, B)\n\n R = np.where(np.logical_and(B_r == 1, B_c == 1), RBgr_BBRR, R)\n B = np.where(np.logical_and(R_r == 1, R_c == 1), RBgr_BBRR, B)\n\n del RBg_RBBR, RBg_BRRB, RBgr_BBRR, R_r, R_c, B_r, B_c\n\n #return tstack((R, G, B))\n return cv2.merge([R, G, B])" }, { "path": "utils/syn/modules/__init__.py", "content": "from .Demosaicing_malvar2004 import demosaicing_CFA_Bayer_Malvar2004\nimport pyximport; pyximport.install()\nfrom .tone_mapping_cython import CRF_Map_Cython, ICRF_Map_Cython" }, { "path": "utils/syn/modules/masks.py", "content": "# -*- coding: utf-8 -*-\n\"\"\"\nBayer CFA Masks\n===============\n\n*Bayer* CFA (Colour Filter Array) masks generation.\n\"\"\"\n\nfrom __future__ import division, unicode_literals\n\nimport numpy as np\n\n__author__ = 'Colour Developers'\n__copyright__ = 'Copyright (C) 2015-2018 - Colour Developers'\n__license__ = 'New BSD License - http://opensource.org/licenses/BSD-3-Clause'\n__maintainer__ = 'Colour Developers'\n__email__ = 'colour-science@googlegroups.com'\n__status__ = 'Production'\n\n__all__ = ['masks_CFA_Bayer']\n\n\ndef masks_CFA_Bayer(shape, pattern='RGGB'):\n \"\"\"\n Returns the *Bayer* CFA red, green and blue masks for given pattern.\n\n Parameters\n ----------\n shape : array_like\n Dimensions of the *Bayer* CFA.\n pattern : unicode, optional\n **{'RGGB', 'BGGR', 'GRBG', 'GBRG'}**,\n Arrangement of the colour filters on the pixel array.\n\n Returns\n -------\n tuple\n *Bayer* CFA red, green and blue masks.\n\n Examples\n --------\n >>> from pprint import pprint\n >>> shape = (3, 3)\n >>> pprint(masks_CFA_Bayer(shape))\n (array([[ True, False, True],\n [False, False, False],\n [ True, False, True]], dtype=bool),\n array([[False, True, False],\n [ True, False, True],\n [False, True, False]], dtype=bool),\n array([[False, False, False],\n [False, True, False],\n [False, False, False]], dtype=bool))\n >>> pprint(masks_CFA_Bayer(shape, 'BGGR'))\n (array([[False, False, False],\n [False, True, False],\n [False, False, False]], dtype=bool),\n array([[False, True, False],\n [ True, False, True],\n [False, True, False]], dtype=bool),\n array([[ True, False, True],\n [False, False, False],\n [ True, False, True]], dtype=bool))\n \"\"\"\n\n pattern = pattern.upper()\n\n channels = dict((channel, np.zeros(shape)) for channel in 'RGB')\n for channel, (y, x) in zip(pattern, [(0, 0), (0, 1), (1, 0), (1, 1)]):\n channels[channel][y::2, x::2] = 1\n\n return tuple(channels[c].astype(bool) for c in 'RGB')\n" }, { "path": "utils/syn/modules/tone_mapping_cython.pyx", "content": "# Power by Zongsheng Yue 2019-06-11 21:23:27\n\nimport numpy as np\nfrom math import floor\n\ndef CRF_Map_Cython(double[:, :, :] img, double[:] I, double[:] B):\n cdef Py_ssize_t h = img.shape[0]\n cdef Py_ssize_t w = img.shape[1]\n cdef Py_ssize_t c = img.shape[2]\n cdef Py_ssize_t bin = I.shape[0]\n\n cdef int ii, jj, cc, start_bin, b, index\n out = np.zeros((h,w,c), dtype=np.float64)\n cdef double[:,:,:] out_view = out\n\n cdef double tiny_bin = 9.7656e-04 # 1/1024 = 9.7656e-04\n cdef double min_tiny_bin = 0.0039\n cdef temp, tempB, comp1, comp2\n\n for ii in range(h):\n for jj in range(w):\n for cc in range(c):\n temp = img[ii, jj, cc]\n start_bin = 1\n if temp > min_tiny_bin:\n start_bin = floor(temp / tiny_bin - 1)\n for b in range(start_bin, bin):\n tempI = I[b]\n if tempI >= temp:\n index = b\n if index > 1:\n comp1 = tempI - temp\n comp2 = temp - I[index - 1]\n if comp2 < comp1:\n index -= 1\n out_view[ii, jj, cc] = B[index]\n break\n return out\n\ndef ICRF_Map_Cython(double[:, :, :] img, double[:] invI, double[:] invB):\n cdef Py_ssize_t h = img.shape[0]\n cdef Py_ssize_t w = img.shape[1]\n cdef Py_ssize_t c = img.shape[2]\n cdef Py_ssize_t bin = invI.shape[0]\n\n cdef int ii, jj, cc, start_bin, b, index\n out = np.zeros((h,w,c), dtype=np.float64)\n cdef double[:,:,:] out_view = out\n\n cdef double tiny_bin = 9.7656e-04 # 1/1024 = 9.7656e-04\n cdef double min_tiny_bin = 0.0039\n cdef temp, tempB, comp1, comp2\n\n for ii in range(h):\n for jj in range(w):\n for cc in range(c):\n temp = img[ii, jj, cc]\n start_bin = 1\n if temp > min_tiny_bin:\n start_bin = floor(temp / tiny_bin - 1)\n for b in range(start_bin, bin):\n tempB = invB[b]\n if tempB >= temp:\n index = b\n if index > 1:\n comp1 = tempB - temp\n comp2 = temp - invB[index - 1]\n if comp2 < comp1:\n index -= 1\n out_view[ii, jj, cc] = invI[index]\n break\n return out\n\n" } ]